Dancing in the Dark: Post-trade Anonymity, Liquidity and Informed

Transcription

1 Dancing in the Dark: Post-trade Anonymity, Liquidity and Informed Trading Alexandra Hachmeister / Dirk Schiereck Current Draft: December 2006 Abstract: We analyze the impact of post-trade anonymity on liquidity and informed trading in an order driven market. The German stock market introduced the Central Counterparty (CCP) in March 2003 for German equities traded on its anonymous electronic trading platform Xetra leading to a major change in its existing transparency regime. Before the introduction trader IDs were revealed to the counterparties of a trade, with the introduction of the CCP even after the transaction the traders remain anonymous. Previous theoretical and empirical research documents that pre-trade anonymity results in increased liquidity, while results on post-trade anonymity are mixed. We find a significant increase in liquidity measured through a reduction of 25% in implicit transaction costs. We also document that the arrival rate of informed traders is reduced in the anonymous setting. Following recent findings of Bloomfield et al. (2005) that informed traders take on the role of liquidity providers we interpret our findings as indication that informed traders change their behavior in providing liquidity more aggressively in an anonymous environment. Jel Classification: G14 Keywords: anonymity, liquidity, information based trading Endowed Chair for Banking and Finance, European Business School (ebs), International University, Schloss Reichartshausen, D Oestrich-Winkel/ Germany; Endowed Chair for Banking and Finance, European Business School (ebs), International University, Schloss Reichartshausen, D Oestrich-Winkel/ Germany; Corresponding author s dirk.schiereck@ebs.de

2 Dancing in the Dark: Post-trade Anonymity, Liquidity and Informed Trading Current Draft: December 2006 Abstract: We analyze the impact of post-trade anonymity on liquidity and informed trading in an order driven market. The German stock market introduced the Central Counterparty (CCP) in March 2003 for German equities traded on its anonymous electronic trading platform Xetra leading to a major change in its existing transparency regime. Before the introduction trader IDs were revealed to the counterparties of a trade, with the introduction of the CCP even after the transaction the traders remain anonymous. Previous theoretical and empirical research documents that pre-trade anonymity results in increased liquidity, while results on post-trade anonymity are mixed. We find a significant increase in liquidity measured through a reduction of 25% in implicit transaction costs. We also document that the arrival rate of informed traders is reduced in the anonymous setting. Following recent findings of Bloomfield et al. (2005) that informed traders take on the role of liquidity providers we interpret our findings as indication that informed traders change their behavior in providing liquidity more aggressively in an anonymous environment. Jel Classification: G14 Keywords: anonymity, liquidity, information based trading 2

3 1. Introduction Electronic trading has become the dominant form of securities trading, allowing for a wide variety of market design setups, i.e. trading model type, priority rules, opening and closing procedures and transparency regimes. The individual market setups follow different sets of objectives, however liquidity being the ultimate goal. Implementing a specific market design may have direct implications on the price discovery process through its impact on trader s strategies and willingness to supply liquidity. Issues connected to different transparency regimes have been discussed widely among practitioners, regulators and researchers. To put the different results into perspective the term transparency has to be clearly defined. Transparency covers three dimensions: pre-trade transparency (order book information, i.e. limits and volumes), post-trade transparency (trade information, i.e. prices and volumes) and anonymity (information on trader IDs either pre- or post-trade). While the positive affect on liquidity of displaying the whole order book (pre-trade transparency) is well accepted, immediate dissemination of trade information (post-trade transparency) is discussed more openly. 1 When arguing the affects of anonymity on trading the underlying assumption is that trader IDs are informative. In a non-anonymous setup all details including counterparty IDs are passed between the parties of the trade. Institutional traders have voiced their concern about front running possibilities and information leakage leading to price impact. In addi- 1 The controversial discussion concerning post-trade transparency was mainly conducted with respect to delayed trade reporting on the London Stock Exchange. While most continental regulators favored immediate publication of trade data the British regulator did not, which was a major issue in the drafting of the European Union s Investment Services Directive as implemented in The effects were analyzed by Naik et al. (1999) and Saporta et al. (1999). 3

4 tion intermediaries use this information to identify trading interest 2 and resort to offexchange trading. This would have a direct implication for order book liquidity and order priority, and of course commercial implications for the operator of the trading system. 3 The German exchange based electronic trading system Xetra provides extensive pre-trade transparency, displaying the whole order book to all market paticipants, however the trader IDs are not displayed in the order book foreseeing pre-trade anonymity. In addition, all transactions with their respective prices and volumes are instantaneously reported providing full post-trade transparency. 4 In a phased approach 2003 the transparency regime was switched and the Central Counterparty (CCP) was introduced for equities traded on its trading platforms (Xetra and Frankfurt Stock Exchange (FSE) Floor) on 27 March and 10 April The introduction was a major change to its initial post-trade anonymity regime. Before the introduction of the CCP post-trading was non-anonymous, i.e. the traders IDs were displayed to the counterparties of the transaction. With the CCP post-trade ano- 2 Market participants have openly stated that in the case of so called iceberg orders in the limit order book, traders tried to hit them with the intention of identifying the trading institution behind the large order. 3 Wells (2000) in his survey of the FIBV members analyses among other market design elements the three dimensions of transparency stating that pre-trade transparency i.e. the distribution of order book information is now common for exchanges with the upstairs market remaining opaque. Post-trade transparency has been discussed more controversial as dealers preferred delaying the information on trades; however limitations are allowed in Europe. For the third dimension anonymity FIBV members supported pre-trade anonymity while post trade-anonymity showed mixed results. Although there was concern by institutional traders for front-running and information leakage only a number of exchanges had moved to post-trade anonymity. 4 The complete pre-trade and post-trade information is provided to data vendors allowing for further distribution. 5 A detailed description of the German equity market is provided in Theissen (2003b). 4

5 nymity was implemented. This change is a natural experiment and offers the opportunity to analyse the consequences of post-trade anonymity on liquidity and informed trading. Our analysis is based on two related strands of literature, the analysis of different transparency or anonymity regimes and informed trading in limit order markets. General theoretical notion for an increase in anonymity based on an inverse relation between adverse selection and informed trading would be as follows: Anonymity is preferred by informed traders, thus an increase in anonymity will lead to an increase in informed traders transactions. As a consequence the adverse selection risk increases and spreads should widen, i.e. liquidity decreases. However, empirical and theoretical results vary. Foucault et al. (2004) analyze the change in the pre-trade anonymity regime for the CAC40 instruments at Paris Bourse. In a first step they develop a theoretical model where in an anonymous setting both informed and uninformed traders bid more aggressively. In a second step they empirically confirm that the introduction of pre-trade anonymity has a positive affect on liquidity. Comerton-Forde et al. (2005) find supporting results for the Tokyo Stock Exchange and the Korea Stock Exchange i.e., an increase in pre-trade anonymity increases liquidity, while displaying broker identities leads to a decrease in liquidity. 6 In an experimental analysis concerning the effects of pre-trade anonymity Perotti, Rindi (2005) show that disclosure of trader s identities reduces the incentive to acquire information, and as a consequence reduces liquidity and volatility. 6 Simaan et al. (2003) argue based on a collusion rationale that pre-trade anonymity reduces the probability for collusion; consequently spreads are lower leading to an increase in liquidity. We believe that the collusion argument will not hold in a pure limit order book market with a large number of intermediaries as for the DAX instruments on Xetra. In addition, a central feature of pure order book markets is the absence of dedicated market makers. 5

6 The literature concerning pre-trade anonymity is fairly consistent and posits a positive relation between pre-trade anonymity and liquidity, which contrasts to the outlined general theoretical notion. 7 Waisburd (2003) concludes for post-trade anonymity in the Paris bourse that bid-ask spreads are tighter when counterparties are disclosed to the market. He compares full non-anonymity (all traders get all counterparty information) to partial nonanonymity where only counterparties of a trade receive the information on trader IDs, but does not include full anonymity in his analysis. His findings support a negative relation between post-trade anonymity and liquidity and contrast with the findings of Foucault et al. (2004) for pre-trade anonymity. Our analysis differs to the analysis of Waisburd (2003): First, Xetra switched from partial non-anonymity to full anonymity where no trader ID s were displayed at all. Second, we look at liquid instruments only, as the initial level of liquidity in an instrument might also play a role concerning the effects of switching the anonymity regime. The differentiation criteria for the two anonymity levels at Paris Bourse were based on the liquidity level (less liquid or liquid). Broad evidence supports the assumption that market anonymity and informed trading are positively correlated, i.e. with the introduction of post-trade anonymity the probability for transactions with an informed trader should increase. Theissen (2003a) investigates the effects of pre-trade anonymity in a specialist system where the non-anonymous environ- 7 Effects to changes in pre-trade transparency may differ from the effects of changes in the degree of anonymity in a market. Boehmer et al. (2005) studying the introduction of pre-trade transparency at the NYSE while holding anonymity constant document improved liquidity along with increased pre-trade transparency (distribution of order book information). In contrast, Madhavan et al. (2005) report that the introduction of pre-trade transparency on the Toronto Stock Exchange led to a decrease in liquidity. From an empirical point of view the results concerning the effects of pre-trade transparency on market quality are less clear. 6

7 ment allows the identification of informed traders thus reducing the adverse selection problem. He concludes that increased anonymity should lead to an increase in informed trading. 8 Frey, Grammig (2005) analyze adverse selection and liquidity in the Xetra trading system. They find a strong positive correlation among spreads and adverse selection supporting the initial hypothesis of an inverse relationship of liquidity and informed trading for limit order markets. In contrast, Perotti, Rindi (2005) document in an experimental pretrade analysis that revealing trader IDs leads to a reduction in informed traders and to a reduction in liquidity implying a positive relation between informed traders and liquidity, i.e. informed traders providing liquidity. Bloomfield et al. (2005) also find contrasting results for trading strategies and order type choice of informed and uninformed traders in an experimental setting. Informed traders are both liquidity takers and providers depending on the state of the order book and the current volatility. They demonstrate that informed traders use more limit orders than uninformed liquidity traders while their market making role emerges due to the fact that they are least subject to adverse selection when providing liquidity. These results contrast with the common assumption that informed traders only consume liquidity through trading strategies with market orders. Our paper investigates if the level of anonymity in post-trading has an affect on liquidity and informed trading. It contributes to the empirical literature in two ways. First, it allows analyzing the change to a post-trade anonymity regime held all other aspects of transparency constant. This unique setup allows results that can be directly related to the changes in the level of anonymity. Second, we contribute to the growing literature on informed 8 These results are also supported by Grammig et al. (2001), who compare the probability of informed trading for the Floor and fully automated trading system IBIS of Frankfurt Stock Exchange. They conclude that informed traders have a clear preference for anonymous markets. 7

8 trading in limit order book markets providing further evidence on the recently controversially discussed relation of informed trading and liquidity. Following the argumentation of Foucault et al. (2004) 9 while assuming that trader ID s are informative not only in the case of orders (pre-trade) but also in the case of trades (post-trade) we conclude that in the anonymous setting liquidity should increase. Hypothesis 1: the introduction of post-trade anonymity has a positive effect on market liquidity: i) spread width declines, ii) order book depth increases, thus iii) overall market liquidity will increase. 10 As an anonymous environment allows informed traders to profit from their private information without being detected, the introduction of post-trade-anonymity should increase the level of informed trading. Hypothesis 2: The introduction of post-trade anonymity leads to an increase in informed trading Foucault et al. (2004) test their theoretical model for the CAC40, the largest and most actively traded French instruments. They assume that the fraction of informed traders is small which is in line with Easley et al. (1996) who find that large and actively traded firms have a lower probability of informed trading. The outcome of their model depends on the size of the fraction of informed traders. For their setting a positive relation between anonymity and liquidity is tested as significant. Our analysis is based on the DAX30 instruments which we expect to shows similar characteristics as the CAC40. In addition, if trader ID s are informative for orders there is no obvious reason why they should not be for trades. 10 In order to test this hypothesis we will compare the results for a liquidity measure covering the different dimensions of liquidity for a defined period before and after the event. The Null Hypothesis tests that anonymity has no effect on liquidity. In order to reject the Null Hypothesis we would expect to find significant differences in means and medians for the pre and post event period for the defined liquidity measure. These results are further validated by a multivariate analysis. 11 We test against the Null Hypothesis that the difference in the mean and median in share of informed traders for the pre and post event period is not significantly different from zero. These results are also further validated by a multivariate analysis. 8

9 Our sample includes data for the thirty German blue chip stocks constituting the DAX 12 over a four month period (February to May 2003). We implement the Exchange Liquidity Measure (XLM) as developed by Deutsche Boerse AG, which measures the costs of a roundtrip in basis points for a defined order volume in. This enables us to determine pretrade liquidity including the full order book depth. 13 For the discussion of informed trading and post-trade anonymity we compute the well known structural model developed by Easley et al. (1996), in the following referred to as EKOP. We test our hypothesis performing univariate and multivariate analysis and document a significant increase in liquidity measured through a reduction of 25% in implicit transactions. The results concerning informed trading are contrary to our hypothesis. The arrival rate of informed traders is significantly reduced in the anonymous setting while the information environment remained constant. As EKOP measures the arrival rates of liquidity takers and not providers it does not capture liquidity provision of informed traders. Following recent findings of Bloomfield et al. (2005) that informed traders take on the role of liquidity providers we presume that the significant increase in liquidity could be associated to informed traders changing behavior in providing liquidity more aggressively in an anonymous environment. The rest of the study is structured as follows. Section 2 gives an introduction to the Xetra trading system and first results for the DAX 30 instruments. Section 3 describes the data set used, the measures applied to identify changes in liquidity and informed trading and the 12 The index DAX is based on prices generated in the electronic trading system Xetra and measures the performance of the Prime Standard s thirty largest German companies in terms of order book volume and market capitalization. 13 Irvine et al. (2000) are the first to suggest liquidity measures based on round trip costs. Gomber, Schweickert (2002) provide an introduction to the XLM. 9

10 overall research design. Section 4 provides the results of the univariate and regression analysis. Section 5 concludes. 2. Xetra limit order book Equity trading in Germany is fragmented between eight exchanges with the Frankfurt Stock Exchange (FSE) playing a dominant role. Deutsche Boerse AG as operator of the FSE provides two separate trading platforms for equities, Xetra and FSE floor. We concentrate on the fully electronic trading system Xetra 14 only, which during the analyzed time period in 2003 had a market share in average traded volume of DAX instruments of around 97%. During the period of analysis trading took place between 9.00 a.m. and 8.00 p.m. CET. 15 The trading model for DAX equities is continuous trading with auctions, i.e. continuous trading starts with an opening auction, is interrupted by two intraday auctions and ends with a closing auction. Continuous trading starts again after completion of the intraday auctions. The trading model is purely order driven, as no designated market makers for the DAX instruments are implemented. 16 Orders are executed in accordance with price time 14 The design and trading rules of the fully electronic trading system Xetra resemble other open limit order book markets such as Euronext as described in Foucault et al. (2004), the Hong Kong Stock Exchange as described in Ahn et al. (2001) and the Australian Stock Exchange as described in Cao et al. (2004). 15 The trading hours were reduced on 1 November 2003 to 5.30 p.m. after consultation with market participants and Exchange Council for the Xetra trading system only. The floor of the Frankfurt Stock Exchange is still open until 8 p.m. 16 In general, provision of additional liquidity through designated market makers so called Designated Sponsors is supported. Market Participants that act as Designated Sponsors can enter quotes into the system. In 10

11 priority. During continuous trading no crossing within the spread is possible, i.e. no price improvement, but marketable orders 17 are allowed to walk up or down the book with no restrictions. Order amendment rules foresee that order modifications lead to a change in time priority if either the limit is modified or the modification has a negative impact on other orders already standing in the book, e.g. volume increase. The basic order types (market, limit, market-to-limit) can be specified further through execution conditions, validity constraints and trading restrictions. The system also allows for iceberg orders, which are from a market design perspective designed to enable traders to provide liquidity when they do not want to reveal the full size of their orders. The hidden quantity of an iceberg order loses its priority to visible quantities at the same limit. Orders entering the order book can be executed full, partially or not at all, thus generating one or more trades or none at all. The Xetra system has also implemented trading safeguards in auctions and during continuous trading in order to improve price continuity and increase the execution probability of market orders. These are volatility interruptions and market order interruptions. Volatility interruptions are triggered if the potential execution price lies outside a defined static or dynamic price corridor around a reference price. Market order interruptions do not play a order to be traded in the trading model continuous trading with auctions an instrument requires at least one Designated Sponsor. However DAX instruments which have sufficient liquidity as measured by the Exchange Liquidity Measure and their trading volume are exempted from this rule. Designated Sponsors are in contrast to Specialist or Market Makers not obliged to provide quotes. Their quoting performance is measured daily and a ranking is published quarterly. 17 Marketable orders are either market or limit orders with limit prices that make them immediately executable. 11

12 role in DAX instruments. The instrument tick size is defined as with a minimum order size equal to the minimum tradable unit for DAX instruments, i.e. one. The Xetra market model 19 also defines the level of transparency through the type and the extent of information available to market participants during trading hours: All orders and prices in the limit order book and all transactions with volume and price are immediately distributed to the trading members (pre-trade and post-transparency). The transparency level can be defined as quite high. However, trading is anonymous as trader IDs for orders are concealed. With the introduction of the CCP anonymity was extended to the post-trade layer. The CCP was introduced with a phased approach: the change for the first set of instruments was introduced on 27 March 2003 and for the second set of instruments on 10 April 2003 with DAX instruments included in both instrument sets. The major advantages of the introduction were the implementation of post-trade anonymity, the reduction in counterparty risk, optimized collateral management and improved settlement efficiency through netting. The CCP is the counterpart of any trade on Xetra, allowing for netting of all transactions on member and instrument level. Our analysis focuses on the affect of post- 18 In contrast to other exchanges e.g. Euronext, the tick size for equities traded in Xetra is not a function of the stock price level. 19 The ultimate and legally binding terms for trading at the Frankfurt Stock Exchange are laid down in the rules and regulations of the exchange, especially the Exchange Rules and the Terms and Conditions for Transactions. The market model serves as a basis for the rules and regulations which, nevertheless, may contain additional terms. In addition to the Market Model for Equity trading, Xetra provides a Block Crossing model and an additional model for warrant and bond trading ( Continuous Auctions ). For details concerning the market model see Deutsche Boerse AG (2004): Market Model Equities - Xetra Release

13 trade anonymity on informed trading and liquidity. The other positive affects of the CCP are not discussed in the context of this study. 20 Table 1 reports for the cross section of instruments of the DAX means and standard deviations for average daily volume in million, average daily trades, average price and average volatility. 21 Volatility is measured as logarithm of the ratio of highest and lowest price during each 30-minute trading interval of the trading day. Market capitalization is reported as of 28 February 2003 and 31 May 2003, with respective weighting factor (index weight) calculated based on the free float factor, number of shares and actual closing price. The results show that the market for blue chip stocks is active. Averaged across stocks 7,400 trades are executed per day with an average volume of 22,800 per trade including partial executions. The statistics also document that the sample instruments differ with respect to trading activity and market capitalization. The number of trades of the most actively traded stock is eleven times as high as the instrument with the least trading activity. The same holds true for amount traded (most active is 80 times as high as least active) and market capitalization (largest firm is almost 60 times as large as the smallest). 3. Data and Research Design 20 Central Counterparties have become popular not only for derivatives markets, where due to the high risks involved they are common for decades but also for equities. For US equities the NSCC (National Securities Clearing Corporation) is the defined central counterparty and was already introduced in It should be noted that the parameters for trading activity count both sides of the transaction. For parameters on order book turnover a simple division by to two is required. However this still includes partial executions of orders. 13

14 The data used in this study are obtained from the StatistiX database provided by Deutsche Boerse AG for a time period of four month from 1 February to 31 May The CCP was introduced in a two step approach for all instruments. The dataset for our analysis was reduced to reflect this approach. Due to the technical complexity of the implementation for all market participants we drop one week prior to the phased introduction. In addition we exempt the trading days between the introduction of the first and second set of instruments. After these treatments our data set contains 30 instruments and 66 trading days with 33 trading days before (3 February to 19 March 2003) and 33 trading days after the event (11 April 30 May 2003). To measure liquidity we use the Exchange Liquidity Measure (XLM) as developed and implemented by Deutsche Boerse AG. The XLM was introduced in July 2002 in order to provide its market participants with the ability to identify the implicit transaction costs, to determine the trading parameters on Xetra as well as the assessment of measures to improve its market model. It is a measure assessing the liquidity of the limit order book on the basis of implicit transaction costs. 23 It follows that, the higher the liquidity the lower 22 StatistiX provided three separate files for all trading days during February to May 2003 for the 30 instruments constituting the DAX: Trades file, BBA file, XLM file. The trades files contains all time stamped trades with related order information i.e. order number and order entry time stamp. The BBA file provides the best bid and asks during the trading day and the XLM file included intraday values for all available volume classes. 23 The XLM covers three dimensions of liquidity: market breadth, market depth and immediacy in execution. The fourth dimension market resilience can be assessed through the change in results over the course of time. The XLM thus follows the concept of four dimensions of liquidity as described in Harris (1990). Gomber et al. (2005) analyze the time dimension of liquidity resilience and find that large trans- 14

15 the implicit transaction costs (measured as market impact) and vice versa. Market impact is a measure of the costs for immediate demand for liquidity. It is the sum of the liquidity premium (LP) measured as the half bid-ask-spread 24 and the adverse price movement (APM) measured as price effect if the order size exceeds best bid-ask-size; both are separately calculated for each side of the book. Adding these components results in the XLM 25 ; this is the measure for the costs of a roundtrip for a defined execution volume. XLM calculates the market impact costs for a given transaction volume in basis points based upon order book snapshots taken every minute from the Xetra order book. These snapshots also include the hidden volumes of iceberg orders providing a measure for committed rather than visible liquidity. The XLM is computed for eight different volume classes (in thousands): 25, 50, 100, 250, 500, 1000, 2000 and 3000 for DAX instruments. For the analysis three volume classes (50, 250 and 500) were chosen. 26 The XLM assesses the liquidity of actions are timed, i.e. discretionary traders with large orders wait until the liquidity in the market is high enough. 24 The liquidity premium added for both sides of the order book is the well known and often applied quoted percentage spread. 25 The XLM provides the weighted average price at which an order of a given size can be executed immediately at time t with PB,t(V) and PS,t(V) where the index (B,S) indicates if the transaction was buyer- or seller initiated and V denotes the order size and MQt the quote midpoint at time t: PB, t ( V ) MQt XLM B, t ( V ) = 10,000 MQt and MQt PS, t ( V ) XLM S, t ( V ) = 10,000 MQt which leads to XLM t ( V ) = XLM B, t ( V ) + XLM S, t ( V ). 26 The transaction sizes > 500k were excluded from the sample as they are far from any typical order size. In addition 25k was excluded as in this case the liquidity premium (LP) would be much larger than the adverse price movement (APM) and as the quoted percentage spread is included as a liquidity measure no additional information is provided. 15

16 an instrument and does not provide a direct measure for the importance of adverse selection 27 although a part of the implicit transaction costs calculated by the XLM is based on adverse selection. When interpreting the results of the LP, APM and XLM it should be taken into account that these measures are pre-trade liquidity measured based on the current state of the order book. They do not give any information about the type of liquidity provider, i.e. if uninformed liquidity traders or informed traders 28 provide the liquidity and who provides what fraction of the order book liquidity. In order to estimate the importance of adverse selection or informed trading in an instrument we compute a structural model (EKOP). Although EKOP was initially developed and tested for specialist markets, it has been implemented successfully for open limit order books. 29 This underlies the assumption that limit order traders behave like designated dealers or specialists when confronted with the possibility of informed traders. 30 The EKOP 27 Adverse selection costs are part of costs imposed through the bid-ask spread (usually inventory, transaction and adverse selection components of the spread are distinguished). As such the XLM includes the adverse selection costs as part of the market impact calculation, but does not provide any information on their share of the market impact costs. 28 We follow Bloomfield et al. (2005) in assuming that liquidity is also provided by informed traders. 29 The EKOP model has been applied to order book markets for example by Brockman, Chung (2000), Grammig et al. (2000) and Grammig et al. (2001). 30 Kyle (1985) concluded that a model with one market maker compared to several competitive market makers lead to the same results under a zero profit assumption. Thus, in the limit order book the sum of all liquidity providers resembles the market makers. Supporting this assumption Chung et al. (2004) find that almost 75% of the bid and ask quotes on the NYSE reflect interest of limit order traders on at least one side of the book. It should be noted that market maker inventory does not enter the EKOP model, thus there is 16

17 model is estimated as follows: Assume that prior to the beginning of each trading day nature selects that an information event will occur (with probability α) or not (with probability 1-α). With probability δ the information will be a bad news event, with probability 1-δ it will be a good news event. Uninformed traders arrive each day with probability ε. Informed traders only arrive on days with information events at the rate µ. Assuming independence across days, which is consistent with an informational efficient market, a direct estimate of α, δ, ε and µ is obtained using the maximum likelihood function from EKOP. 31 Then, the probability of informed trading (PINF) is π = αµ/(αµ+2ε). The data input required for the maximum likelihood function is the number of buyer and seller initiated trades per trading day. Computing these statistics is straightforward on automated order driven markets, as trades can only take place at the posted bid and ask prices. Per definition price improvement is not possible. The trade initiator of any trade can be identified through the available order entry and execution time stamps for all orders involved in the trade. In addition, there is no need for an algorithm to identify partial executions of large orders, i.e. consolidating trades during a certain time interval. Trades that no difference between market makers and liquidity providers in the costs they are facing, i.e. adverse selection and transaction costs. In addition, the EKOP model assumes only for liquidity takers (informed and uninformed) exogenous arrival rates while an exogenous arrival rate for liquidity providers is not necessary to ensure model consistency. 31 We implement the log likelihood function as presented in Easley et al. (2002): T T T µ S M µ B M B+ S M L({ y t } t = 1 θ ) = ( 2ε + M ln x + ( B + S)ln( µ + ε )) + (ln( α(1 δ ) e x + αδe x + (1 α ) x ) t = 1 t = 1 where M min(b,s) + max(b,s)/2 and x (ε /µ+ε) [0,1]. 17

18 walk up or down the book and generate partial trades are identified and classified as one trade in accordance to their order number. 32 It should be noted that EKOP only measures the arrival rates of market participants that take liquidity (trade initiators) and not of those providing it. In this respect it does not provide any information on the liquidity providers, i.e. if they are informed or uninformed traders and their share in liquidity provision. As stated by Bloomfield et al. (2005) informed traders also act as liquidity providers while the PINF only captures the liquidity taking informed traders and will bias the overall presence of informed traders. Before calculating the different measures data cleansing was done. The described measures for liquidity and informed trading need a best bid and ask limit and quantity for calculation purposes. The unambiguous trade initiator for the EKOP model can only be computed during continuous trading. That means all transactions that took place during an auction 33 either the officially scheduled open, intraday and closing auction or any volatility interruption are excluded from the sample. During the time period of our analysis only 3% of the transactions took place during auctions. However these accounted for 7% of the traded volume, indicating that the average order size was larger in auctions than during continuous trading. During the period of analysis there was no change in the constituting instruments of the DAX, i.e. for all 30 instruments a full data set is available. Descriptive statistics will be provided in the cross section of the instruments constituting the DAX for trad- 32 Xetra attaches an order number to each order entered into the system. No matter how often an order is partially executed the information that the partials belong to one order can be identified with the unique order number. 33 During any type of auction the order book displays an indicative price (if available) instead of the best bidask. In addition, it is not possible to determine the trade initiator during batch auctions. 18

19 ing activity measured as average number of trades per day and the average amount traded in million, volatility defined as logarithm of the ratio of highest and lowest price during each 30-minute trading interval of the trading day averaged across the day 34, average price and market capitalization in million. For the univariate analysis all variables are averaged across the day in order to generate one observation per instrument per day. These observations are averaged for the pre event and post event periods and compared using paired t-tests and Wilcoxon signed rank tests. The purpose is to identify, if the difference in means and medians is zero. If the results are significantly different from zero we can reject the hypothesis that the introduction of posttrade anonymity had no effect on the parameters analyzed, i.e. it did not have any impact on the liquidity in the market or the share of informed trading. In order to test for Hypothesis 1 we compute the statistics for quoted percentage spread (LP), APM and XLM for all three liquidity classes. The multivariate analysis is computed to further underpin the results of the univariate analysis. The regression analysis allows to measure if the switch to anonymity affected the dependent variables, i.e. the different liquidity measures by controlling for other potentially relevant variables. We estimate separate regressions for the quoted percentage spread (LP), APM and XLM for the three different volumes classes. Given well documented relationships between spreads and trading volume, stock price and volatility, we implement these three variables as control variables in addition to a dummy representing either the pre event or the post event period, i.e. the anonymity level, in the following regression: Liquidity = α + β1tradedvol + β2vola + β3 Pr ice + β4perioddummy 34 Volatility it high it = ln lowit 19

20 where Liquidity as dependent variable is defined as either quoted percentage spread (LP), APM or XLM for the three different volume classes. TradedVol is the average daily traded volume in million and expected to yield a negative relation; Vola is defined as the average daily volatility measured as logarithm of the ratio of highest price and lowest price during each 30-minute trading interval of the trading day. We expect a positive relation to the liquidity measures; Price is the average trade price expected to yield a negative relation; PeriodDummy is a dummy variable taking on the value 0 prior to the event and 1 otherwise and should exhibit a negative relation. All variables are calculated for each instrument and each day and are aggregated to one observation per instrument per period. We estimate the regression using standard OLS estimation based on the log regression equation to ensure normal distribution of parameters. To test Hypothesis 2 we calculate the estimates of the EKOP model and perform a paired t- test and a Wilcoxon signed rank test for the complete sample of instruments only. As EKOP is not calculated as a daily variable but as one observation for each period analyzed, i.e. one observation for each pre event and post event period, the paired test cannot be computed per instrument but only for the complete DAX sample which provides 30 observations for both periods. Given documented relationships between volatility, the rate of uninformed traders (substitute for traded volume) and informed traders, we implement the two parameters as control variables in addition to the dummy variable. We estimate the following regression equation 35 : 35 We estimated two regression equations differing with respect to the inclusion of TradedVol as independent variable. As expected we find collinearity between the arrival rate of uninformed investors (UnInformed) and average traded volume. Including TradedVol does not improve the results of the regression. The results are not reported but are available upon request. 20

21 InformedTradingi = α + β1uninformed + β2perioddummy + β3vola where InformedTrading is the arrival rate UnInformed is the arrival rate of uninformed investors (); PeriodDummy is a dummy variable taking on the value 0 prior to the event and 1 otherwise and Vola is measured as the logarithm of the ratio of the highest and lowest price during a 30-minute interval. For all three parameters we expect a positive relation to the arrival rate of informed traders. All variables are calculated for each instrument and each day, leading to one observation for each instrument and each trading day. We estimate the regression using standard OLS estimation on the log regression equation to ensure normal distribution of parameters. 4. Results Hypothesis 1 claims a positive relationship between the introduction of post-trade anonymity and market liquidity. We test for a decrease in order book width measured as a decrease in quoted percentage spread (LP), an increase in order book depth as a decrease in APM and an increase in overall liquidity as a reduction in XLM. Table 2 reports summary statistics for the cross section of DAX instruments; means, standard deviations and differences for the pre event period defined as 3 February to 19 March 2003 and the post event period defined as 11 April to 30 May T-tests and Wilcoxon signed rank tests examine if the difference in mean and medians are equal to zero. The test statistics show that the decline in LP, APM and XLM are significant at the level for the complete sample. On average the quoted percentage spread (LP) declined by 5.9 basis points, which is a reduction of 24.5 % in implicit trading costs. The APM and XLM for the 50k volume class decreased accordingly. The decrease in APM and XLM for the 250k and 500k volume class is even 21

22 stronger, leading to a reduction of 30 to 40 % in basis points. The strong decrease is driven by the decrease in APM as the XLM is the sum of LP and APM and the LP is per definition independent of the volume class, i.e. the same for all volume classes. Although all individual instruments show the same statistical significant reduction in LP at the level, the results for the APM and XLM are mixed. The reduction of APM for the 50k volume class is for 23 instruments significant at least at the 0.05 level 36 while the reduction in XLM is significant at least at the 0.05 level for 29 instruments (or 28 instruments based on the results of the Wilcoxon test). The reduction in APM for the 250 and 500k volume classes is significant at least at the 0.05 level with the exception of Henkel, RWE and Volkswagen, which remained stable during the period of analysis. 37 The same holds true for the reduction in XLM for the 250 and 500k volume classes is significant for all instruments at least at the 0.05 level but with only two exception: Volkswagen and RWE. No negative impact from the introduction of post-trade anonymity on the liquidity in these instruments is found. The results suggest that liquidity increased after the introduction of post-trade anonymity. We do not reject Hypothesis 1 and its sub-hypothesis but find that the introduction of posttrade anonymity has a positive effect on spread width, order book depth and thus on overall market liquidity. With the introduction of post-trade anonymity liquidity increased at the spread and in the order book which leads to the conclusion that more market participants are willing to provide liquidity. These results contrast to the findings of Waisburd (2003) on post-trade anonymity but are in line with the results of Foucault et al. (2004) and 36 For 18 out of 30 instruments significant results at the 0.01 level are reported. 37 For brevity reasons table 4 only reports the APM for the 250k volume class. Results for the 500k volume class are comparable. 22

23 Comerton-Forde et al. (2005) for the effects of pre-trade anonymity on liquidity. However, we do not know who the additional liquidity providers are. As a sideline we computed for the pre event period a ranking for the cross-section of instruments for average daily trades, average daily trading volume in Mio., market capitalization in million and XLM for 50k and 250k volume classes. Trading volume and number of trades are often used as proxies for the liquidity of an instrument. However, the ranking for these variables varies, which leads to the conclusion that the most actively traded stocks are not necessarily the most liquid instruments and trading parameters when used as proxies for liquidity should be used with care. Table 3 reports the results of the ranking. 38 The reduction in quoted percentage spread (LP), APM and XLM documented through the univariate analysis may be caused by variables not controlled for. As the reported parameters of Table 1 also show significant changes between the two periods analyzed we control for the possible effects on our results for the different liquidity measures. As already outlined liquidity is a function of price, volume and volatility. Previous research shows that traded volume and price are inversely related to spread (liquidity), while volatility has a positive relation. 39 Table 4 presents the results of the regression analysis for the complete sample of DAX instruments If the introduction of post-trade anonymity is associated with changes to liquidity we would expect to find the regression coefficients to be significantly different from zero. The coefficients of the controlling variables volume and volatility show the ex- 38 Bayer AG ranked as number 10 in terms of traded volume, 5 in terms of number of trades is only number 13 ranked in terms of liquidity as measured by XLM. 39 See, for example, Stoll (2000). 23

24 pected sign and are significant at the level for all liquidity measures. Price as a controlling variable shows mixed results, negative significant results as expected at the level for APM 50k and XLM 50k but insignificant results for all other liquidity measures. Focusing on our binary indicator for pre the event and post event period, the beta coefficient shows a negative sign which is significant at the level for LP and APM 50k. For all other liquidity measures results are not significant. The results for price and our binary indicator change to highly significant levels with expected sign for all liquidity measures when the independent variable volatility is excluded from the regression we find a strong correlation of over 0.5 between volatility and both parameters, leading to collinearity problems. The adjusted R squares of over 90 % and the highly significant F- Statistics show that the different regressions fit the data rather well. In summary the multivariate test implies that the introduction of post-trade anonymity lead to an increase in liquidity, supporting the results of the univariate analysis. We continue with the analysis for Hypothesis 2 which proposes a positive relationship between post-trade anonymity and the level of informed trading. Table 5 reports the results of the estimation of the maximum likelihood function of EKOP. All parameters are calculated for the pre event period and the post event period. As in EKOP (1996) we restrict and to (1,0) and and to (0,). The probability of informed trading (PINF) is computed as the function of the model parameters as The probability of an information event () for the combined sample is 0.23 during the pre event period, ranging from 0.06 to 0.38 suggesting that private information is released about once a week. The probability for an information event with an average of 0.29 () for the combined sample ranging from 0.05 to 0.64 is not significantly different for the post event period as Wilcoxon test results show. The individual results of our sample do not support the general notion that 24

25 instruments with a higher trading volume are associated with more information events, then the instruments with lower trading volume. Implementing sub-samples based on average traded volume does not yield different results. 40 However as our sample is constituted of the 30 largest and most liquid German blue ship stocks we would not expect a pronounced difference. The probability of a bad news event () for the combined sample is 0.57 with 0.19 and 0.88 as low and high during the pre event period. Results for the post event period with 0.48 for the combined sample and 0.10 and 0.99 as low and high do not significantly deviate from the pre event period as the results of the Wilcoxon test show. For both periods the probability for either good news or bad news is almost equally distributed. The order arrival rate for informed investors () during the pre event period is 90.0 ranging from 25.8 to 290.6, while it is significantly reduced to 71.9 during the post event period. Wilcoxon test results show significance at the 0.05 level. In contrast, the order arrival rate for uninformed investors () does not show any significant difference between the pre and post event period. The arrival rate for the complete sample is for the pre event period and for the post event period. To determine a firm s level of informed trading EKOP takes the relation of arrival rates of both traders and the probability of an information event in account: The PINF value for the combined sample is with 0.04 as low and 0.14 as high value during the pre event period. During the post event period the PINF is significantly reduced to ranging from 0.04 to The reduction 40 See table 3. 25